Latest / Notion in Practice / Lawyer to Ops Lead: Building an Agent-Powered Startup | George Turnbull, Jerwin Parker, Tim Jeffries
Transcript
- Tim Jeffries: So George is doing a lot. She's building. She's cooking. That's very cool. Actually how, like how deep she's gone building operational agents that really are leveraging. Jerwin Parker: Cheers. Tim Jeffries: Like all the capability, you know, what are they being out? Two weeks or something? Jerwin Parker: Mm-hmm. Welcome everyone to this episode of the Nurture in Practice podcast. Today we have the amazing George joining us. She is an ex-lawyer turned into the head of operations at Calcs and George, welcome. weeks and she built like the AskOps in what five minutes? ⁓ Slack AskOps that's cool. ⁓ Tim Jeffries: Yeah, it's so good. So good, so good. Yes. takeaways, mate, what did you think? What are your three? George: Thanks, Joan. Thanks, Tim. It's good to be here to talk about one of my favorite products, Notion. Jerwin Parker: Yeah, first one for me is going on the leverage part of the Ask Ops agent she built. Like the first key takeaway is that can create like real ops leverage and by turning ⁓ conversations that they're having on Slack into actions and that sort of governance piece. Like ⁓ quoted and I wrote it down, ⁓ ⁓ built it to keep people where they work on Slack. ⁓ Tim Jeffries: Yeah. Jerwin Parker: the system's still in Notion. And so I thought that was pretty cool. What do you think, Tim? Tim Jeffries: Yeah. Yeah, yeah, no, I had one really similar, like you build the agents to work with people, like where they are. I feel like the capability for that should expand over time. So at the moment, you know, there's this really neat connection with Slack. and ⁓ think for a lot of people, it's kind of the missing ⁓ for Notion. Like there's no asynchronous chat, you know, in the tool. Jerwin Parker: Welcome. ⁓ Tim Jeffries: I have to apologize straight off the bat. sound like I'm 300 years old today. I don't think the AI can fix that, can it, Joen? No, right. Jerwin Parker: no. So just address it to the viewers. Why do you sound different today, Tim? Tim Jeffries: It's called a head cold and there's no AI fix. So it's just life. Apologies. Other than comments, which is not really built for that. So it makes a lot of sense to have agents like interface with Slack really nicely. But yeah, the fact that their company works so much on Slack, will allow them to leverage. Like notion in a, in a much stronger way. And interesting that the company's only been, what'd she say? Six, seven months. ⁓ a huge amount of time. ⁓ Interesting though. ⁓ Jerwin Parker: There could be an AI fix. can turn up the tone of your voice to very high or low pitch up to you. But back the conversation, George. So you're an ex-lawyer. I'd really like to know about how you, ⁓ one, we can into Notion, but your extensive career history. Cause it's not every day I meet someone that ⁓ used to be lawyer and now is the head of operations at a scaling company. So yeah. Yeah, six to seven months. Yeah. Tim Jeffries: When it comes to the cost piece, they are going to be paying for tokens for their team to stay in Slack and leverage Notion. Whereas if their team were in Notion, they would be able to do a lot of things that the agent is ⁓ for them by talking to their personal agent in Notion, which they wouldn't pay for. So interesting like how that will. Jerwin Parker: Could you tell us a little bit about your journey? George: Yeah, I suppose it's an unusual career path. Most of my mates from law school are still practicing law. I only practiced for a short time. I was a family lawyer and then a commercial litigator. And then I moved into a legal tech firm. So I suppose that was my first introduction to tech and I really liked it. So I stayed. I worked prior to Calcs at a company called Linktree where we Notion extensively and now heading up ops ⁓ at ⁓ Jerwin Parker: Mm-hmm. Tim Jeffries: impact the cost for them. Jerwin Parker: Yeah, they're already thinking about it and they're staying lean and flexible towards it, I think is super important to be able to just experiment and see what works. ⁓ of like our intro, right, is the Lego blocks, like play around ⁓ with Lego blocks and eventually you can build something really cool. But ⁓ I think that goes to the next takeaway, which is like always Nice. And how did you end up at using Notion originally? Was that at Linktree? George: Yeah, so in 2020, when I joined at Linktree, we just introduced a cool new product called Notion. I think it was our CEO, Alex, who got us on board with the tool. And really ⁓ my kind of brief as the new operations associate from a legal background to operations associate was to figure out Notion and how we can use it at Linktree. And so I really... Tim Jeffries: Mm. Jerwin Parker: work on the foundations, like when you build a nice Lego piece, I'm to use Lego as an example, you always work on the foundations first. Like you look at a blueprint document, which is sort of like, there's a context layer and then there's how do we keep all these pieces clean together? So like all the blue blocks will go here and all the red blocks will go there. hopefully that makes sense to him. but had a really strong George: worked alongside Notion as we scaled Linktree from, think, when I joined a 40 person company to around 150, 160 or so. Yeah. Jerwin Parker: Wow. Tim Jeffries: Yeah, right. And the early days then too. So, I mean, there was a bunch of things in the product that are not what they are today and performance wasn't what it is today. Like how was it in those early days? George: yeah, I mean, was reflecting on that even just yesterday as I was building a Notion agent, it was much more, I suppose, one dimensional in terms of knowledge management. And there wasn't the kind of like system to action that's available in the product today. There was a lot of manual work to get people to adopt ways of working in Notion that can now be automated ⁓ or ⁓ using Notion AI. So yeah, certainly the product is Tim Jeffries: Yeah, you mean build it up from the foundation first. She talked a lot about that. Yeah. Jerwin Parker: Exactly. Exactly. Yeah. She talked about doing that and then she had that hygiene, was an agent and then talking through the verification stuff was really cool. Tim Jeffries: Yeah. Yeah. I mean, that was another takeaway for me was just, always in these systems, the quality of the information, the accuracy of it, the truthfulness of it, like directly impacts the overall usefulness of the system, how much people trust it, how effective it is. That's always been true, but ⁓ had to like keep that just by George: really evolved, but I think like some of those like foundational building blocks that I relied on at Linktree are still what I rely on today to make those enhancements like Notion AI work better and faster and more efficiently. Jerwin Parker: Mm-hmm. Mm-hmm. Tim Jeffries: Yeah, I'm finding that a lot. Everybody would love just to go straight to the AI piece, but unless you have that structure in place, it just doesn't work. And so yeah, for a lot of companies who've never done that sort of thinking before and organizing, they sort of have to go back and do that first. But the payoff is much higher than it used to be. It used to be good to do that, good to be organized, ⁓ helpful for the team, but ⁓ you couldn't get the return that you can get now. brute force before, like lots of ops people spent ages making sure that databases were like the content in them was accurate. You know, like if new things got added, did they have these properties? And you could do it a little bit with automations and you could create sort of audit views to try and surface anything that was a problem. But she's really taken this line that's gone, Hey, we can get agents to keep this information up to date. And, you know, she was just saying now the workspace is really humming. George: Absolutely. Jerwin Parker: Mm-hmm. George: that's absolutely right. ⁓ really about building that context layer, that foundational piece, those building blocks. And that's allowed us to unlock huge efficiencies ⁓ Notion AI. Tim Jeffries: You know, and I think what we'll see is the tool useful increase. People will trust it more because they know that the information is accurate because the agents are doing their job. So yeah, I thought that that was really useful. And then the last takeaway for me is really that like now is the time for experimentation. It's all moving so fast. We just got to build and it's going to be messy. And if an, an ops person can, can allow that, then everyone should get on board. Jerwin Parker: Nice, nice. And did you take a lot of the blueprints and principles that, move from Linktree and then, ⁓ moved it across to Calcs? ⁓ was that process like? Yes. George: Yeah, I tried not to be too set in my ways. I have revisited some of those foundational pieces and actually worked really closely with Tim on building out a foundational layer that's much more built around core company databases as our warehouse, our knowledge management warehouse. And then it's about building operational leverage off those. So Tim Jeffries: Right. Normally those people will be trying to bring sort of structure and you can see, you know, her need for that. certainly feel that too. ⁓ you know, wish that there was an orchestration layer, wish that we had more visibility, you know, wish that the notifications were better, but like it's all being built. we just got to get in there and play with it and, experiment and then, you know, let the dust will settle. And we've got this window, right. Until May where, ⁓ the cost piece we don't have to have completely resolved yet. So if there's a time to play it's now. Jerwin Parker: course. George: having that layer of hygiene and knowing that every document, I think where we got to at Linktree without that hygiene layer is that there were sort of abandoned documents here and there in places, whereas at Calc I'm trying to be really disciplined with the help of agents to make sure no document doesn't have a home. And there's a really clear hierarchy and hygiene as to how we sort and organize our documents, which has been a real game changer. Jerwin Parker: Yeah. Exactly. upon that key takeaways is that is where we're headed next, right? It's ⁓ she more agents, ⁓ the turn into like, ⁓ she mentioned it herself, like ⁓ how mentioned the agent org chart ⁓ and sort you know, the examples of like having visibility of what the agents are actually doing, better ways to get like failure notifications and Yes. Tim Jeffries: Mm. Are you using the verification piece in Notion? Yeah. George: Yeah, we are. So all of our documents in the databases owners and they're verified. And then we have hygiene agents that go in and on a weekly basis check the documents after date, check last date. So there's like a threshold requirement. ⁓ If they been edited in a certain time, the agent will go in and check ⁓ and ask assigned owner to re-verify. Jerwin Parker: the accuracy gap we also discussed as well between ⁓ tools to comparison to having it all ⁓ within ⁓ And of course the forced thinking now of the usage-based pricing, forces you to now be more concise and lean with your spend. Post-May, I think that's what ⁓ we'll all thinking because I'm already thinking about that. I think I mentioned you. ⁓ Mm-hmm. Tim Jeffries: Yeah, nice. George: ⁓ We're also using the verify function really effectively in our software and tools. And actually our CEO, Chris, came up with this process, but we have an agent who requires you to verify that a tool is still in use, that a renewal date is still current. So that we don't go over on software and tools that we're not under on or over or under on licensing. So the relevant tool owner required to verify that that's set up for that tool software. Jerwin Parker: That's really cool. Tim Jeffries: Yeah. Yeah. Jerwin Parker: chat of our notion baddies chat that was like, I've ⁓ been to put some of my notion agents on ⁓ performance ⁓ because the credits I spent like, I don't know how many obviously ⁓ you earlier. I was like, I've spent so much time ⁓ building these random different agents. And then I checked the credits. was like, Whoa, Whoa, Whoa, Whoa, Whoa. That's a lot of money that I've been spending. Well, not really spending, but if I was spending it, I was like, ⁓ no. Yes. Tim Jeffries: Yeah. Hahaha George: ⁓ remains true on a monthly basis, which is actually for a small company like us, really important because software and tools can easily spend a lot of money there and we want to make sure that we're spending money effectively. Tim Jeffries: Yeah. Yeah, nice. I have a little, saw a little sneak peek of like, improvement to the verification tool, in a job I was working on recently where you're to be able to mark things as archived. so just like you can verify them, you can archive them ⁓ and ⁓ it that can sort of stay in place, but they won't appear in the search and they won't appear to the AI. So the AI won't like utilize them and ⁓ index them. Cause Jerwin Parker: I've got to cut that down and be more strategic in what I build and versus what I actually do manually with my notion AI chat agent versus a custom agent. George: ⁓ yeah. Tim Jeffries: Yeah. Yeah. It's going to be interesting once we switch across, cause I think the experimenting, well, it won't be free anymore. Right. And so we won't be able to just be curious and be pushing it around. We'll have to, ⁓ yeah, consider it first and work at return on investment, all that sort of stuff. So anyway, that was great. I really enjoyed that. ⁓ ⁓ yeah, we'll look forward to the next one. Jerwin Parker: Mmm. Tim Jeffries: problem about sort of company knowledge where you need to keep something for like, you know, long-term sort of awareness, but what do you do? Do you pick it up and move it out of the structure that it's in, in which case then it sort of becomes hard to find. you know, ⁓ it's not connected to its team or whatever anymore. but don't want AI like referencing the old information that's in it. ⁓ yeah, I think. Jerwin Parker: Mm-hmm. George: Mm-hmm. Jerwin Parker: Yeah. Tim Jeffries: I think we might see more of that, like that content status property, you know, has verify and we've had that for a little while, but then once it has archive too, then you can see these kinds of knowledge management things, both with the agents and then also a feature like that. Yeah, I just, it's going to go up a level. think. Sure. Well, that's a wrap for this episode. Please leave us a comment. Let us know how this is useful. If you've got ideas about other topics that we can cover, angles, if you want to come and be a guest, reach out, let us know. We look forward to seeing you in the next episode. George: Yeah, yeah, and build context because I think like then you can when you're chatting with Notion AI too, they can say like, hey, we have tried that before. Like that sort of thing is part of the work history and part of the iteration, which is really important too as you grow. Tim Jeffries: Yeah, for sure. Yeah, cool. So you're building agents then. You're making them. Lots of people are sort of just starting, but it sounds like you're well and truly in, did you get them early? Did they show up in your workspace before? George: Yeah, mean, yeah, I mean, I've tried across a couple of different platforms. I think super straightforward. Like I was chatting to a couple of friends who in tech and telling them about what I'm building in Notion and they were like, wow, can you give me a crash course? I was kind of like, just get started. Like it's pretty straightforward, natural language prompting. like, yeah, it it really accessible to build an agent. Like if I tell my mom I'm building an agent, she... This is like shocked, thinks I'm a genius. ⁓ But it's pretty straightforward. I think like the key challenge I've had with Notion is like going from knowledge management, pretty one dimensional, probably like the sort of five years ago when I was working with Notion. And it was a bit of a push to get people into Notion to search for the right documents. Now we're finding like operational leverage through Jerwin Parker: Yeah Tim Jeffries: Yep, yep. Jerwin Parker: Mm-hmm. George: agents taking that knowledge management to action. So one of the that I built in like five minutes the other day was actually based off a template, just ⁓ really basic Q &A template that is essentially like an AskOps bot available now across our Slack workspace. And so at any time when you call AskOps ⁓ or if a question frame that the agent recognizes is an ops related question, it's So one of the agents that I built in like five the other day was actually based off a template, just ⁓ a really basic Q &A template that essentially like an AskOps bot available now across our Slack workspace. And so at any time when you call AskOps ⁓ or if there's a question frame that the agent is an ops related question, it's Jerwin Parker: Wow. Wow. George: answers that question in the thread and then also stores that information or pinpoints where there's a missing people policy or artifact that will answer that question. answers that question in the thread and then also stores that information or pinpoints where there's a missing people policy or artifact that will answer that question. Jerwin Parker: Mm-hmm. Mm-hmm. George: And also like helpfully compiles how many times that question has been answered so that then we can say, we really need like a better expenses policy. It's obviously not clear. ⁓ So that kind of like organizational alignment that you get from like there's less back and forth and I'm like an ops team of two. Tim Jeffries: Mmm. George: I've come from a much bigger ops team. don't have HR, we don't have like a talent team. So I do all of that. And my goal is like, we don't actually need to add more heads. I can do a lot of this through operational leverage I get from Notion AI. So we've got like a TA system that we've built out with an agent. We've got onboarding facilitated by an agent in ⁓ Notion. yeah, there's lots of examples I can touch on, but it's just like finding that like, extra bandwidth for me through these agents. Tim Jeffries: It is real, isn't it? Like it's actually, I feel like for the first time stuff's doing it now too, ⁓ regular people building agents that are genuinely able to amplify their like effectiveness. I'm doing a job with a company and they're like, we, we're a 10 person company. We want to appear to the world, like a hundred person company. And that sort of sounded ridiculous a little while ago, but now I'm like, yeah. Jerwin Parker: Mm-hmm. Tim Jeffries: I can see how we can do that. You know, when you look into people's roles and what they spend their time doing, you can amplify one piece of work because either the agent can take the thing that they doing and sort of do it repeatedly, or they can be managing a series of those agents that are kind of doing those tasks. It's, yeah, it is a little crazy and people who are not doing it, like a long way behind, you know, without kind of being crazy about it. Like it's very quickly. Yeah. I'm just, I don't know for the long time I was, I was a bit of a cynic. I wanted it to work because I'm a nerd and I thought it was cool, but I, but I wasn't seeing it in practice. You know, I was just seeing hype and YouTube videos of people saying, Hey, you can do this cool stuff. But knowing that it was just the realm of the nerds, but it's not anymore. Jerwin Parker: Hehehe. George: Totally. Yeah. It's it's a new way of thinking too, cause now I go into any task and my default is there's an agent for this. Like there's a way to do this faster and better while still that repeatable quality. Like our, our of AI who's, who's very experienced. ⁓ We're obviously on AI product internally. And he said these days reading Sorry, writing is cheap, reading is expensive. And so I think one of the like real caveats here is like in terms of like actual document generation. So there's the agent stuff. Then there's like a lot of the document generation stuff that happens with Notion AI at Calcs as well. We've obviously got access to like Google Docs, Slack, Notion, Linear. And so we can produce good work that knows how we work and what we're up to. We want to make sure it's really accurate and really quality. So I think that's the big caveat with like Jerwin Parker: Mm-hmm. George: just letting these things happen. Like there needs to be guardrails and parameters around how it's used. And that's probably what we're grappling with as well now. Tim Jeffries: Yeah. How are you finding the notion tools on that? Because I feel like it's all happened so fast that they're a bit thin. I'm, I'm seeing things like custom instructions. Every person creates their own version of that. And if you've got 40 people or a hundred people or whatever, like there are things that you would want those custom instructions to, you know, be the same across the whole company, but there's no mechanism for that beyond, I know doing something with a sync block and asking people to use it or something like that, you know? George: Yeah. Yeah. Totally. Yeah. And I think that's where like being at a small company, I don't have too much of that. Like it's me and Vanessa, our ops person based out of Columbia. And so there's not too much overlap, but I can see where that happens. What I think is missing or would be a great evolution of the product is for me to be able to visualize and see my agents like on a screen, like who's working together and who's connected to what and what are they doing every day and better notifications. Tim Jeffries: it ⁓ Jerwin Parker: Hmm. Tim Jeffries: Mm. George: around when they fail because sometimes like if you don't have the permissions right, like if they lose access to certain pages and they just stop firing, unless it's like a Slack agent that you'll see the stops responding to the ask request, you may not necessarily necessarily say, ⁓ they've stopped producing the AI summary of the candidates that are coming through. And so we're not getting those sent for screening that should be under that process. probably better, yeah, a better view of the orchestration. think it's on us and, notion AI can help in terms of like, what are our guidelines in terms of, terms of, and ways of working in terms of setting those instructions ⁓ that can, we can apply that, ⁓ as part of an agent itself. but yeah, it's, it's more about like seeing the bigger picture that systems way of thinking. I think though, at this point. Jerwin Parker: Sorry. George: We're a bit loose on too much rigor and structure because we want people to experiment. you know, like have a go, like build an agent. That's awesome. Like even at the end of the day, we've got three different agents doing slightly similar things. That's okay. Everyone's experimenting with the product. ⁓ then we get to a point down the track where we've kind of really codified like how these agents are going to work and work together. Tim Jeffries: Yeah. Yeah. Hmm. Yeah, yeah. I have to ask, have you looked at the cost yet? George: No, No, we have. mean, yeah, we have like, we have, there's like, candidly, I think they've been really clever in the way that like, it's been available to everyone and we've really, ⁓ upped our usage and we're building this reliance on the functionality. So, yeah, obviously need to start charging for that. costs like the usage costs money. so yeah, that'll be interesting. Jerwin Parker: Good question. Tim Jeffries: Yeah. I mean, there's always, when everyone brings out costs, like there's all this backlash, right? It's like everybody wants it for free, but I think, I ⁓ it's good that ⁓ have to face this bit because you know, when we have like the rest of AI inside of Notion is kind of unlimited usage, right? And so you've got no sense of what it's actually costing to do what you're doing. ⁓ and ⁓ people sort of have this angle, they just want to get as much out of it as they can. But the reality with the custom agents, I think, is you are building something that is an investment. It's an investment to build it and get it right as a part of your system. But then it is going to cost to run it. And company is just going to pay that token, ⁓ for free forever. ⁓ then it forces us to think, OK, what is the most effective way to do this? How we do it? Our companies had access to the agents for a little while and I was the only one playing with them. And I think in two weeks I spent like 1500 US dollars, with these agents doing all kinds of crazy things. I was like, I mean, I didn't actually spend it because it's free at the moment, but you can see the cost ⁓ your agents. Yeah. So it's, it's worth looking into. Jerwin Parker: Yeah. George: Yeah. Totally. And I think like you've touched on the positive spin there, which is like, it'll just force more discipline. Like it'll force us to know what agents, ⁓ often they're running all those things. And that'll be, that'll be a good thing. Tim Jeffries: Yeah. Yeah. Yeah, that's right. Yeah. And so how are you going sharing them? Cause one of the things I've been finding a little bit clunky is we set up an agent, I'll set up an agent and then I'll try and share it with someone else in the workspace. How that piece works has a bit of a challenge. George: Yeah, I mean, I'm looking at my sidebar now. So we have like our entire workspace. We just make them freely available to everyone. ⁓ I guess that's kind of what I was getting at before. Like I don't have that like. why do you almost like it would almost be like a, like an order. What's the word I'm like for the org design or chart of agents. Like that's kind of what it be great to like. Tim Jeffries: Yeah, right. George: visualize the org chart because we've got different like functions. So I've got my like HR function agents who look after TA and onboarding. Then we've got all our like database hygiene agents. So attached to each of those databases. I just want to see that all in one place. And then if you could share that view, people can see, okay, what's available currently, what's left to build. And then looking into them seeing what they're doing. I haven't had too much trouble sharing maybe because ⁓ Tim Jeffries: Yeah, yeah. George: We're just like, we're a smaller business and like a small ops team. And, and for now that's it's working. Tim Jeffries: Yep. Yeah, nice. And are you doing anything with the calendar and the mail kind of piece or is it more, it sounds like they're more kind of ops focused. George: Yeah, the one thing that I do sync with, this is still ops focus, but I've created, so like for all of our leadership rituals, we have a leadership calendar and we use fellow for note taking. So we're not actually using notion for note taking. We may get there, but at this point, the integration between fellow and Google mate is too hard to be. So ⁓ we basically any calendar event that comes from a leadership ritual. The fellow notes sink directly into our leadership database. The notion agent then draws out decisions made by leadership into our decisions database and actions for leadership into an actions database. So that's quite nice. Cause I used to spend time like as the sort of operational person in the room, you know, pulling out the actions, pulling out decisions and sharing a recap. And that all just happens automatically with a notion agent now. Tim Jeffries: Yeah. Yeah. Mm. Yeah. Yeah. That's been the biggest change in my workflow. And I think I looked back the other day and I'd had eight meetings pretty much back to back all day. I hadn't stopped after any of them to kind of look at it. And you just, you know what that's like. Like you just lose the context in the moment. think, yeah, yeah, I'm across this. I know what's happening. And I know that there's some notes I can go back and have a look, but can't think about it in the same way as you could right at the end of that call. And the fact that I can just trust, yeah, it pulls the Actions out, it puts a little summary there. It updates the project. Like I know that I, not losing anything here. It's taken a little while for me to build enough trust that it works, but it works. It works better than I used to, to honest. ⁓ George: Yeah. Yeah. still probably in that building trust mode though. Like I'll still go back and check like, this get moved into the database? Like, is it still humming? Cause I do think sometimes, you know, you get that notification in the settings where it's like, it's failed to run or something, a permission, something like better notification, talking product improvement now, but like better notifications of when they're not running. I think it would be a good ad. Tim Jeffries: Yep. Yeah, yeah. I know how your notifications going. I mean, the little red dot in the top left that I, you know, used to religiously try and keep, it's out of control. George: Yeah, yeah, yeah. Yeah, that's what I think. Yeah, it gets it gets really noisy, like some better way to notify that these things are broken even on the agent, like some kind of like, yeah, something on the agent itself. Tim Jeffries: Mmm. Have you got a favorite? George: A favorite agent? Let me think. ⁓ one that I'm building at the moment that I think will be a favorite, but I still haven't really proven ⁓ that, so we've got our, we work mostly, we're a global distributed company, we mostly in Slack. Our Slack workspace gets a bit out of control with channels and things, so. ⁓ Tim Jeffries: Yeah. George: and we've got like a Slack ways of working guide, but you know, extra channels get created, not in keeping with the naming conventions, all those things. So ⁓ the moment I'm doing a big like cull and clean up. These are like the glamorous ops activities. and then I'm going to have an agent look after using the like Slack ways of working convention. An agent will, it's a little bit clunky cause I have to export. a CSV as the trigger, export a CSV of our ⁓ Slack channels, upload it to like, I'll get a notification to upload it monthly to the agent. The agent does a review of that CSV and it tells me which channels should be archived or renamed based on the conventions. So it like, it's not rocket science, but it saves a bit of time because I used to just do that pretty manually. Tim Jeffries: Yeah. Yeah. Totally. Oh, that's good. I thought you were going to tell me you're making a policeman, you know, and it was going to, when it saw someone not obeying the ways of working document, it was going to wander into tell them. Really? Yep. George: You know, that's a Slack, that's actually a Slack AI function. So we don't, we don't pay for Slack AI. So there's some workarounds. I think one of the challenges we have as a small company or any company that discipline with their software and tools is like, we don't want this AI sprawl. Like that would be easier to do in Slack. There's things that I'm trying to do in Notion AI. Like we run our whole like TA, like we get, you as a We use Zapier to bring in a form when someone applies for a role at our career site. The rest of it is managed by a Notion agent in terms of like tracking that applicant's history, setting up buttons for when they through the stages. All of the communication with them is captured in their candidate profile. We could easily pay for a HR, I don't TA solution that is built for that. Jerwin Parker: on us. George: But the challenge and like the cool thing about a small business is that like we're building these things with the existing tooling we have. So you have to be a bit creative. And I actually find that like pretty interesting. Tim Jeffries: Yeah, it is interesting. And it compounds like having that context all in the one spot. going, doing what I do, ⁓ see so often like a team, just, they won't be able to leverage AI on a process because there's bits of it in multiple systems. As soon as, you know, half of that lives in, I don't know, you know, a different tool. The AI can't see the whole thing, doesn't have enough context, can't really be useful. ⁓ but if you deal with the George: Boom. Tim Jeffries: clunkiness of using the existing tooling and like finding a way to, you know, maybe the notion form isn't quite as good as like the external form, like teleform or, know, whatever else. But like, if you can deal with that bit, then you get end to end flow inside of notion. yeah, you can, you can deal with that. And the agents really help. They have smoothed a bunch of the, you know, previous manual, yeah, issues. Yeah, that's good. George: Yeah, yeah, totally building that context. One of the other things that I'm working on at the moment is like, we've obviously got the ask ops agent that you can tag into Slack. And I think we probably like a lot of the way that I work is like, how do we keep people like people are mostly working in Slack. So like, let's make it so that they don't have to necessarily go into Notion. So they tag at ask ops and they ask about operational related things. We'll have the equivalent then for, you know, ask development, ask engineering. Jerwin Parker: Mm-hmm. George: ⁓ ask marketing, it just frees up marketing time. If Notion can scan Notion, Google Docs, Linear and tell us where things are at. And working through the logic at the moment now in terms of also triaging requests through that same process. So ask marketing, I need a marketing page ⁓ up in the next two days. It automatically gets triaged and added to their projects list. That kind of thing, keeping all of that. in one space and having one consistent process across the business. That's something I'm working towards. Like at the moment, the ask function is there, but it should also be the work request. Because if we keep our projects and tasks in ⁓ then we should be initiating those requests with Notion as well. Tim Jeffries: Yeah. Yeah. Yeah, that's cool. My team, a little bit kicking and screaming, but we dragged them into Slack so that we could like play with this integration. And now basically every project, you know, we have a workflow that, ⁓ that notion agents can't do it yet, but we use NADN to spin up like a, you know, once a project hits a certain stage, it gets its own Slack channel created. and then George: Nice. Tim Jeffries: It puts the ID for that Slack channel back in the project in Notion and a link to it. So it's kind of easy to jump between the worlds, but all the async chat between sort of, you know, the consultant and the backend guys or whatever about that project all happens inside that channel. And then ⁓ I think at the moment I've got it set every four hours, I might change that once we start paying for it, ⁓ it goes and has a look at that channel and sees, ⁓ hang on. This conversation has progressed. And it brings that context back to Notion. So now our Notion project pages, they just have all of the context about a job. So they have the meetings and the tasks and, you know, like a summary of the Slack conversation. They have all our emails, like when we send emails out, the bot looks at who it was sent to, looks in our people database, sees that that person is connected to that project and puts that context on the project. So now there's this one place. George: ⁓ huh. That's great. Tim Jeffries: our world ⁓ all of the context about the project comes into. And I'm certainly finding that it knows more than ⁓ do, because it's caught up that, hey, those two guys were having this chat about fixing this problem. And no one has to remember to make a note about that anymore. It's just captured. George: Yeah. Cool, alright. Yeah, might have to borrow that one from you. Tim Jeffries: Yeah. Yeah, yeah, absolutely. I know that it's kind of fun, right? They're like a new version of sort of templates, I think. can't ⁓ them as easy, but like conceptually you pick it up and say, yeah, okay, I like that idea. I'm going to find a way to, yeah, build context that way or do whatever. George: One of the, just thought one of the challenges that we have had is just, cause we've got obviously like when you've got software developers working out of linear in our case, a lot of like the day to day. And I think we spoke about this team when we were setting up our motion workspace and a of the day to day of the non-software dev, non-engineering in our case in notion tracking projects and things like a few times we like OKRs to we score OKRs every quarter. Tim Jeffries: Mm. Mm. George: to align the team on strategy. And a few times ⁓ looked to automate progress updates on OKRs. sometimes, you've got to be really deliberate in your instructions of like, basically unless you know this to be true, don't say it. Because I've had a few instances where ⁓ it'll take from ⁓ linear and software developers will be like, hang on, ⁓ that's not right. Tim Jeffries: Yeah. George: So I don't know if that's a context thing or tweaking of the instructions thing, but that's just like an issue we've had. Tim Jeffries: Mm. Yeah, I think there's a, there's a gap between when the content lives natively in Notion, what you can see, like what, what it, AI's understanding of it is very true, right? Cause it's like, it's, it's right there in Notion. uses the relations. It's not doing so much of the big semantic searching to try and find what it's looking for. It follows the relation and it goes, ⁓ here's the content. Whereas when it goes out to these external services, the further out George: Yeah. ⁓ Yeah. Tim Jeffries: it has to go to get them the less like, direct context it's getting. And so. Yeah. I actually wrote something, that I think Joe and we can give, give to people this week that is really about like, how does the notion AI get its context? Like when it says it can connect to SharePoint, what does that really mean? And unfortunately you can't put a SharePoint link in and then the AI follows the SharePoint link and then goes and looks at the document on the other side. It doesn't work like that. George: Yeah. Tim Jeffries: It does like a semantic search across your SharePoint, looking for that information, but you don't have as much control about what it finds. Whereas when you have the relations to say, ⁓ you know, this project is connected to this OKR. It doesn't go looking for what project am I trying to work out, you know, the status of all the tasks on or whatever it just connects to it directly. So it'd be interesting to interrogate that linear piece and work out how is it connecting the linear projects to the OKRs. George: Hmm. Jerwin Parker: you George: Yeah, and I wonder if the first cut of that would be like having better project pages like the ones that you described where all of that relevant project information is first piped into that project page relevant, relevant, like, like from the Slack channel as well as linear all the different places where the work happens. Jerwin Parker: Mm-hmm. George: Because yeah, unhelpfully, it's not just Slack and Notion for us. It's also linear in a lot of cases. that was then consolidated in one page. That becomes the reference point for those like high level ⁓ progress updates relevant to that project. Yeah, there's definitely a bit of tweaking there, but we've come a long way from having distributed systems that the software devs are manually updating in various places. Tim Jeffries: Yep. Yep. Yep. It's... ⁓ yeah. Yeah. Yeah. Yeah. We had systems that before looked so great on paper, but in practice just didn't get used enough. I just, was never disciplined enough to, you know, go and put in the client engagements. I chatted to them on the phone or whatever. Whereas now it all, you know, just happens ⁓ automatically. It's funny. I go into these databases that beforehand were a good idea, but didn't really get used very much. And when I go in there, all of the entries are the bots. George: Exactly. Yeah, yeah. Tim Jeffries: It's not the humans putting anything in. I put one in myself this morning and I was like, ⁓ it's novel. know, just cause it does such a good job, but so consistent, ⁓ know, tracking and capturing that context. George: Really? Yeah. ⁓ Yeah, I think familiarizing the team for us like with because sometimes like especially we migrated over from ClickUp when I about when I started so we're only sort of like six, seven months into using Notion as a company. Probably should have said that at the beginning. But we came over from ⁓ and so there was like a set way of working and people are getting used to Tim Jeffries: Mmm. Jerwin Parker: Mm-hmm. George: working with Notion and there's a few questions around like, where should this document go? And adjusting to the databases can be like a learning curve. And it's just getting used to the idea that you can just ask Notion AI, like, hey, where does like copy paste, where should this stuff go? And like, it knows how Notion ways of working, it'll tell you where to go, it'll add it to the database. So yeah, just familiarizing people with that potential and way of thinking. Tim Jeffries: Yeah. Yeah. That's the promise end state, I reckon is that people have their little agent that understands, you know, their world and the context so well that it's actually all day, every day, kind of helping them along. I think it's hard to get people there. Like it's hard to help them get to that place, especially if they're, if they're not sort of AI. I don't know. I don't want to say AI native, like AI comfortable because George: Yeah. Jerwin Parker: You Tim Jeffries: Like I'm, I'm working with a lot of guys and I find they've, they've sort of adapted to AI, but they use it like a vending machine. You know, it's like here, I'm giving you my thing and then you give me my bit back and then we're done. You know, it's very just like this transaction and I'm trying to encourage them to say, like, think about it more like a relationship. I know that's weird because it's, it's a bot, like talk to it, you know, try and get it to understand what you're, what you're doing. Don't just give it one prompt and say, run actually, you know, go back and forth. George: Mm. Jerwin Parker: Yeah. Yeah. Tim Jeffries: you will build up the context. will understand more and more. It will get clearer and clearer inside this chat and then ask it to go and do what you want it to do. And if it's got instructions about how the workspace works, then it'll do that thing and it'll put it in the right spot for you and it'll be your little helper. That's a big shift for a lot of people that they're really not ready for that or not used to that. George: Mm. Tim Jeffries: Totally possible though. It's pretty cool. Big challenges. Like what are the big next things on the list you reckon that you haven't been quite ready to tackle? George: Probably. Yeah, probably just like the ask ⁓ ⁓ triaging, project triaging from Slack Notion. But then I think it's also, think like ⁓ bigger it is just like, how are we, like, what's the system? Like we've all got a couple of agents that we've built. Like what's the orchestration here? Like I need to do some bigger picture thinking of like, how do these work together? ⁓ Where do we use agents versus like Slackbot versus Zapier versus Claude co-work? Like there's a few things in the mix right now. Like pricing will be one ⁓ to consider and will be forced to be more disciplined. But it's like, what is the, what's the agent org chart for us? ⁓ Yeah. And we're like, ⁓ yeah, starting to formalize our, our understanding. Jerwin Parker: Yeah. Tim Jeffries: Yeah Jerwin Parker: I like that concept. George: But there's a bit of a way to go for that bigger pictures system thinking. Which I think I'm comfortable in the ambiguity at this stage, because I think as ⁓ noted, you're of the curve if you're using agents. So I'm comfortable in the fact that we're experimenting and we're not putting too much pressure on the teams to have all the answers just yet. Tim Jeffries: Yeah. Yeah, nice. Jerwin Parker: Have you started creating like a decision framework as you look at all these tools and processes and ⁓ your AI chart? George: Not really, because I think it'll be a little bit bottoms up. So at the moment, we're testing the potential. ⁓ then we'll step back and decide, OK, yes, ⁓ is our knowledge management tool. makes sense to orchestrate agents from here to this extent. And this is the Notion agent org chart. And there may be supplementary agent tools as well. But yeah, we're still firmly in the like. Jerwin Parker: Mm-hmm. George: trying by doing stage. think like in terms of the low learning curve, we're still like climbing the mountain. And then the product keeps changing and so. Tim Jeffries: Yeah. Do I know that's what I was going to say. Like, do you ever get there? ⁓ Thankfully it's useful on the way up. Like it's just not all learning. Yeah. Jerwin Parker: You George: Yeah, I'm like to be. Yeah, exactly. It's useful on the way up, but like, yeah, to be honest, like it was probably like there are things I'm doing differently now that like I wasn't doing in January. There's things like, like former colleagues at Linktrio like message them and be like, I can't believe we did XYZ manually. Like, can you believe we man like, so we have a of the agents I also really like is our like OKR progress ⁓ So like Tim Jeffries: Yeah. Yeah. Jerwin Parker: Yeah. George: on a fortnightly basis, it just takes out, our scorecard. does a summary of like how many are on track off track. And like, for those that are off track, it goes into a bit more detail. And so then people can thread on like solutions or we've just got that level of oversight. I used to do that manually in a slide deck. Like it's just things like that where it's, you just no longer have to think about, and it just happens. And you can look back and posting the Slack channel yesterday. And I was like, ⁓ that's, that's good. Like that reads well. ⁓ Tim Jeffries: Yeah, right. Jerwin Parker: Mm-hmm. Tim Jeffries: Yeah, yeah. Have you bumped into... It is, yeah, yeah. A bit like having a new colleague. you bumped into where the agent's doing something that you used to do ⁓ ⁓ didn't realize that it had some flow on benefits? Like, you know, maybe something like the OKRs. I coming across these instances where we automate a process with an agent and ⁓ another hole appears because when a human used to do that bit manually, they were catching, you know, some things that were falling through or they saw opportunities because they had a manual process. And this automation of the process sort of, no one realized that that was happening, but has surfaced. ⁓ hang on. If no one's going to like read that document line by line anymore, we're just going to get a summary of it. We're going to miss these, you know, these things. so it's good ultimately overall, cause it sort of surfaces. Jerwin Parker: Mm. Tim Jeffries: holes in the system that people didn't realize they had. But yeah, I'm wondering if you've bumped into those kinds of. George: I think so I'm reflecting on just how manual, how much those like kind of tedious manual tasks took up so much of my job and time. And that switch in thinking of like, there's an agent for this and it forces just ⁓ better. administration and discipline, like the procurement workflow agent that we have where I would, we're usually, we would have like people manually, mean, people still do they manually, well, they prompt the agent in Slack manually fill in the form about the software that they want to buy. But it then automatically, and this is actually using automations in databases as well. then automatically, once it's ⁓ through a Slack workflow automatically gets added to our software and tools database. And that then verify process kicks off. And it's just like those holes where like previously it would have been up to me to at like 11 PM at night be like, ⁓ did I add Figma to the software and tools database? Like, yeah, it's just like freeing up thinking and freeing you up from those like tedious pieces of work that, like tedious pieces of work, but also like, you don't want to miss it. ⁓ there a like, Tim Jeffries: Aye. Yeah. George: There's a real reliance now on agents, I would say. Tim Jeffries: Yeah, yeah. Yes, I know. Well, we start paying for them in May, so we've got till then to get organised. George: I know. Well, they've their wage in a lot of ways already for us. So yeah, we'll see what that looks like. But yeah, to our point earlier, like I think the main change will be we'll get a lot more disciplined. Like I'm staring at a drop down of like multiple, multiple agents, some that are just like practice runs and us mucking around and we'll obviously bring a lot more structure to that. Jerwin Parker: . Yeah. Tim Jeffries: Yep. Yep. did want to ask about external tools. ⁓ ⁓ know, agents can do a bunch of things and we, we often spend a lot of time talking about notion, but, ⁓ I, ⁓ using ⁓ as well. George: Yeah, I've only really just dabbled in Claude. I just did a co-work course. And I've connected to Notion and Google Drive. And so it's got the integrations. I was actually doing stuff in Claude, making changes to a Notion page the other day, which is pretty mind blowing. ⁓ Tim Jeffries: Yep. So using the MCP connection between Claude and Notion. Yep. George: Yeah. I keep thinking like, it's like, who's going to win? Like, where do we, where do we land? Like, and why? I think that's the conundrum I have at the moment. It's like, should I be outside of Notion tinkering with Notion? Is that a better space to be like, or should I be ⁓ Notion? That's where all my knowledge that that feels like where I should be building these agents, given that knowledge hub. ⁓ Yeah, I'm still Tim Jeffries: Mm. Mm. Jerwin Parker: Mm-hmm. George: figuring out. There's certain things that like, just because of the way that we work, like if we want to export a Notion document, we're not going out to customers in Notion document form, we're going out in Google Docs. If we want to export a document that's going to a customer that formats correctly into Google Docs, we've been exploring and Claude works really well for that, that kind of thing. So, but again, like the forcing function here is spend, like, like we don't want to have... Tim Jeffries: Yeah, nice. George: GBT, Claude and Notion agents running. We just need to figure out who wins at the right things and that system, like that orchestration system that we'll land on. Jerwin Parker: Mm-hmm. Tim Jeffries: Mm. Yeah. Yeah. We don't know. George: But yeah, think heavily like in the kind of experimentation phase still with like competing platforms as well. Tim Jeffries: Yeah, I think Notion's quite weak on the external file piece. like, yeah, well just, sorry, even just external, um, file types. like Word docs, for instance, like I had a client I was chatting with yesterday and they just live in Word, you know, mean, so many places are still just full Microsoft houses. And so they're so used to PowerPoints, Word docs, Excel spreadsheets. George: the export. and Jerwin Parker: Mm-hmm. Mm-hmm. George: Hmm Tim Jeffries: but you can't hand any of those to like Notion agent and it can't produce those for you. And so if you want to live your life in Notion and build your systems and run everything in Notion, but you have to send word docs or you have to interrogate spreadsheets, you just can't, right? And something like Claude that does that really, really well. know, I'm big job with a financial services group and you know, everything's in a spreadsheet and the spreadsheet is doing crazy lots of calculations and Jerwin Parker: Yeah. George: Yeah. Yeah. Tim Jeffries: You'd never try and do it in Notion, like it's spreadsheet business, but Claude just talks to the spreadsheet you know, and it's able to produce these insights and reports that in the end we take back to Notion because we want the context, but Notion can't do it. George: Yeah. Yeah. Yeah, that's the kind of mental gymnastics and I'm like, like a lot to figure out in this space. Jerwin Parker: Yeah. Tim Jeffries: ⁓ apologies. ⁓ George: Yeah, is interesting on the notion side it. Like my ⁓ instinct like, if I'm, if all of my knowledge lives in notion, I'm doing the work in notion. I'm asking, ⁓ making requests using notion That just feels like instinctively, Claude just feels too ⁓ separate in of like context building. ⁓ but that, yeah, I guess like there's limitations to both. Tim Jeffries: Far away. Yeah. Yep. Yeah, I think the closeness to the context is huge. Things like, ⁓ be able to cobble the pieces together, like use the database automations to set it up so that when the agent runs, it runs really efficiently. you it then the agent does something and then puts the content that it, ⁓ you back into Notion and then notifies the right person and that piece, like, I feel like that's going to be the moat for ⁓ Notion. George: Hmm. Jerwin Parker: Mm-hmm. Tim Jeffries: is that while your digital operating system is kind of built there, maybe you'll go out to specialist tools, sort of like we always have, to play a specialist function that Notion hasn't built yet or is never going to do. But in the end, you're going to want the context to land back in Notion. It's going to be the sort of central brain. George: Hmm. Hmm. I was thinking the other day actually that Zapier do a really good job of, ⁓ just illustrating like what the agent is doing in those blocks. ⁓ so like, you know, notion agent trigger is Slack, ask ops responses, ⁓ through stored in this database. Like, know, the, if you spend much time in Zapier, I think that's probably what I'm searching for is on like. Tim Jeffries: That's how we're building anyway. Jerwin Parker: Mm-hmm. George: the visualization of what these agents are doing. Sometimes storing it in the instructions can be a bit too... Yeah, yeah, it's not visual enough. Yeah. ⁓ Tim Jeffries: Yeah. Jerwin Parker: visibility piece, yeah. Do you find that's the scary part when you're using Claude? Cause I've done that too, is when you give Claude access to your notion and it asks for us, permission on Claude to update your notion page and it's telling you it's doing it, but you can't see anything that's happening until the output's done. ⁓ you can't really say stop on whatnot. but I like the, your, your insight of like, having that visibility of what George: Yeah Jerwin Parker: the agents actually touching what it's doing specifically because if you give it access to ⁓ or like very, very close to the heart databases that you manage day to day, like that gets deleted. ⁓ It's big issue, you know, cause everything else breaks. ⁓ George: Yeah. Yeah, yeah. Actually, I had this like freak out. I used Claude to read our brand design guidelines ⁓ Canva and then take a proposal written in Notion, apply those brand guidelines and export it as a Google Doc. And then... ⁓ Jerwin Parker: Mm-hmm. George: like the next day I was looking at just like an all hands deck of ours that had just coincidentally also been updated in the brand guidelines. And I thought for a second there, I was like, has Claude gone and just updated our entire workspace? Luckily, no, that was all done separately and deliberately. But it's a bit like that, isn't it? Because like, yeah, ⁓ you see as much in Claude of what, you know, I really like how notion do the strike through in your instructions, you have to accept the changes. ⁓ Jerwin Parker: You ⁓ good, many hana takk. Yes. George: It feels like you're really along for the ride and you still like own the process. Or is clawed sometimes I'm like, well, what have I just unleashed? Jerwin Parker: Yeah. Tim Jeffries: Hmm. Yeah. Yeah, nice. Cool. Well, this is great. Thanks George. So, so much. mean, I'm loving kind of sharing the journey with people cause it's just, there's not enough hours in the day to explore it all yourself. Like I leave every day with a kind of list of things that I wish I'd had more time to play with. listening to other people's experience, I haven't built as many of those operational workflows. A lot of the agent stuff I've been doing has been connecting it with mail and calendar and George: Yeah. Thank you. Jerwin Parker: Yeah, thank you. George: Yeah. Tim Jeffries: you know, building those kinds of workflows for people. So yeah, I'll look forward to the chance to do more of that. George: Yeah, that's interesting. yeah, I think one of the takeaways in your what you're doing with your setup with projects, I'm going to borrow from, because I think like we've got a lot of slack project channels. And because we have people, you know, waking up in Colombia, waking up in Canada, trying to count out what's happened in Australia. That would be I can I can picture how that would work. Jerwin Parker: Mm-hmm. George: So I'm probably going tinker with that all afternoon. As you do, like I think like one of the things I've learned as well is like you can spend quite a bit of time just like playing around with the possibilities, which is fun. It's like a real experimental stage. Tim Jeffries: Yeah. ⁓ Yeah, yeah, it's great. Cool. All right, well, let's do it again in like, I was going to say six months, but probably like three weeks and the whole world will have changed. Jerwin Parker: I'm gonna go. George: Yeah, totally. It's crazy the speed that it's all changing. Tim Jeffries: Yeah, Cool. All right, well thanks so much. George: Awesome. Jerwin Parker: Thank